cooperation, conflict, and the costs of...
TRANSCRIPT
Cooperation, Conflict, and the Costs of Anarchy∗
James D. FearonDepartment of Political Science
Stanford University
Forthcoming in International Organization
August 13, 2017
Abstract
I consider a model in which two states choose how much to arm and whether to attack in successiveperiods. Arms are useful not only for deterrence or taking territory, but also because they influence theresolution of a set of disputed issues. It is shown that states can cooperate on the issues by limitingmilitary competition, but only as far as a “war constraint” allows. Factors determining the tightness ofthe war constraint imply hypotheses about the international determinants of military effort and thus thecosts of anarchy. The strategic logic differs from standard security dilemma arguments, in which thecosts of anarchy are associated with conflict between status quo states that are uncertain about others’territorial revisionism. Here, inefficiency arises because arming to deter lowers a state’s value for livingwith the status quo, which creates a security externality and a feedback loop. The model both synthesizesand revises a diverse range of theoretical arguments about the determinants of interstate cooperation andconflict.
In Theory of International Politics, Kenneth Waltz famously analogized states in international
relations to firms in an oligopolistic industry.1 Firms choose prices (or quantities) in an effort to
maximize profits, but in the limiting case of perfect competition they earn no profit due to a strate-
gic incentive to undercut each other. Given their preferences, anti-trust law, and “the structure of
∗Thanks to Bertel Hansen for research assistance; to Avidit Acharya, Darin Christensen, Charles Glaser,Richard Jordan, Nuno Monteiro, Robert Powell, Ken Scheve, and two anonymous referees for advice andcomments; and to the Canadian Institute for Advanced Research, IOG group, for research support andfeedback. An earlier version of this paper was presented at the 2015 Annual Meetings of the AmericanPolitical Science Association, San Francisco, September 3-6.
1Waltz 1979.
1
the situation,” the firms face what amounts to a Prisoners’ Dilemma. Waltz saw this model as ex-
actly parallel to the idea that international anarchy leads states to choose policies – arming, allying,
and attacking, in particular – that can make them worse off than if they could instead commit them-
selves to cooperate. For Waltz, the microeconomic analogy was a way of developing and putting on
solid ground the traditional Realist intuition that there is something fundamentally different about
interstate relations, something that makes them inevitably tragic in a way that domestic politics
need not be.
Theory of International Politics was written before developments in game theory reworked
the field of industrial organization, and in particular before elaboration of the idea that repeated
interactions could allow firms to construct self-enforcing agreements that would sustain higher
prices and profits.2 Robert Axelrod and Robert Keohane applied this idea to international relations,
suggesting that it undermined the claim that they must necessarily be conflictual and tragic.3 Joseph
Grieco argued in response that if states have an intrinsic preference for “relative gains,” it could be
impossible to sustain a cooperative interstate agreement by implicit threats of ceasing cooperation
if the other cheats.4 Robert Powell argued that the right Realist counter should not be to make a
different assumption about state preferences, but to understand that in anarchy force can enable
states to convert relative advantage into absolute gain.5 Depending on military technology and
2In game theory, the core idea of the “folk theorems” dates to the 1950s, but extensive development withapplication to oligopoly theory occurred in the 1980s. In his first book, Waltz 1959, 191-92, considered butlargely discounted the possibility of getting interstate cooperation from implicit threats of a trade war in hisdiscussion of anarchy and international trade.
3Axelrod 1984; Keohane 1984.4Grieco 1988. Grieco suggested that states’ intrinsic preference for relative gains makes their interactions
more “zero sum,” thus reducing the scope for cooperation. But if Realism is identified with the claim thatinternational politics are zero-sum between states, then a Realist can’t also argue that international politicsare rendered tragic by anarchy. In a zero sum game you can make one party worse off to the benefit of theother, but you can’t make both worse off than they ideally could be.
5Powell 1991.
2
costs for fighting, this possibility might prevent states from cooperating despite a “shadow of the
future” (i.e., repeated interactions). Tragedy restored, but with qualifications.
Parallel to this neoliberal-neorealist exchange, an ongoing debate among Realists centers on a
different objection to the idea that anarchy implies a great deal of tragedy. Robert Jervis argued that
tragic outcomes were due to “the security dilemma,” and that the severity of the dilemma varies
with the degree to which military technology and geography favor offense or defense. Jervis’
core arguments, developed further by a number of scholars taking a similar approach, picture state
leaders as uncertain about whether other states are satisfied with the territorial status quo or have
expansionist preferences. When defense is favored in the “offense-defense balance,” arms levels
can safely be lower and cooperation is less risky if another state is in fact expansionist. Using
this theoretical basis, “defensive realists” maintain that because offense is disadvantaged due to
balancing behavior and because nuclear weapons effectively favor defenders, international anarchy
needn’t or shouldn’t be the source of large costs.6 “Offensive realists” disagree due to reservations
about the offense-defense balance concept and because they think that even a small probability of
facing an expansionist type is enough to make major powers expansionist or defensively aggressive
in turn.7
Routinely taught in graduate International Relations seminars, the arguments referenced above
are interesting and important. They conflict in some places but overlap in others.8 It is not
clear whether or exactly how they might emerge and survive as implications of a common set
of premises.9
6Jervis 1976, 1978; Glaser 1992, 2010; Kydd 1997, 2005; Van Evera 2001.7Brooks 1997; Mearsheimer 2001.8In his constructivist take on the anarchy question, Wendt 1992 pointed out that if states did not under-
stand themselves to have conflicting preferences, they could not be troubled by Prisoners’ Dilemma-typeproblems, and suggested further that repeated cooperation might incline states’ leaders to feel bad aboutacting in ways that disadvantaged their partners. Realists disputed the relevance of both points.
9Wagner 2010, chap. 1.
3
This paper develops a two-state model built on Realist premises. The states in the model
(1) interact in the absence of a third-party that can reliably guarantee agreements between them
(anarchy); (2) can gain from coordinating their interactions but have some degree of conflicting
preferences over issues of mutual concern, which may include territory; (3) have the ability to gen-
erate and use armed force against each other; and (4) find both the provision and use of arms costly,
at least to some extent. The goal is a baseline model that succinctly characterizes central strategic
problems faced by states interacting in international anarchy, and that yields clear implications on
what determines variation in the costs of anarchy over time and across states.
To outline the model, two states choose arms levels and whether to attack in successive peri-
ods. If there is no war, relative military strength determines the resolution of a set of issues through
an unspecified bargaining process. If there is a war, territory is put into play, so that the winner gets
not only control of the issues but also the loser’s territory. I show that repeated interactions can
be used to enable mutually beneficial cooperation between the states on the issues, but subject to
a war constraint: arms levels cannot be so low that a state would prefer to “break out” by rapidly
mobilizing and going to war. Thus, as “neoliberal institutionalists” like Keohane and Axelrod ar-
gued, there can be considerable scope for cooperation despite anarchy due to the possibility of
self-enforcing agreements that rely on repeated interactions. However, in line with the traditional
Realist intuition, this mechanism cannot in general support the first-best outcome of no war and no
costly armament (or other competitive effort). Some costs of anarchy are avoidable without supra-
national government, some are not. At least for the two-state case, the model shows just what
follows from a set of standard premises – that is, states that have the capacity to use force, some
conflict of interest, but also a common interest in avoiding wasteful and dangerous competition.
The baseline model yields a set of propositions on the size of the unavoidable costs of anarchy.
In line with Jervis although due to a different logic, the war constraint is less severe, and thus the
potential depth of interstate cooperation greater, the more geography and military technology favor
4
defense relative to offense. The war constraint is also less tight the less intrinsic value the states
put on controlling the other’s territory, which can be an effect of industrialization, democracy,
or nationalism when agreement on homelands and borders exists. Greater “gains from trade” –
meaning more potential value to be divided up that is separate from homeland territory – may
either increase or decrease the war constraint due to offsetting effects. Protectionism, or any factor
that lowers the efficiency of exploitation of gains from trade between independent states, tightens
the war constraint, raising the arms level necessary for a stable peace. In dyads where one state
has greater military potential than the other, the smaller state is predicted to have a higher arms
burden, but the model reveals offsetting forces that push in the direction of military burdens being
only weakly related to state size.
Although not normally stated in terms of arms as a measure of international cooperation,
some of these implications will seem familiar to IR theory adepts. In particular, some results from
the baseline model look like a synthesis of positions argued by “defensive realism,” “neoclassical
realism” and “neoliberal institutionalism,” with additional implications on conflict and trade remi-
niscent of ideas from, for example, Cordell Hull and Richard Rosecrance.10 Other positions, such
as several associated with offensive realism, are not supported.
What is new then? First, theoretical synthesis. These and other implications emerge in a
connected and transparent way from a single strategic model. This is in contrast to the standard
presentation of arguments given above as competing schools of thought that are based on or stress
different assumptions or issue domains.
Second, the model provides an alternative theoretical foundation, since the logic that generates
the implications differs markedly from standard arguments. The closest thing to a clear baseline
model in the existing literature – and also the theory undergirding defensive realism – is Jervis’
10On Hull, see Schatz 1970, 89; Rosecrance 1987.
5
treatment of the security dilemma and subsequent developments of it.11 There, great power politics
are driven primarily by states’ uncertainty about other states’ preferences for territorial expansion.
By contrast, while it could be added in principle, this model has no uncertainty about types at
all. We still get, however, standard security-dilemma-like results concerning the offense-defense
balance. Indeed, the model suggests a novel way of formalizing the somewhat mysterious idea of
the security dilemma: Even without any uncertainty about states’ preferences, in the model greater
offensive advantage leads to greater equilibrium arms levels, which lower states’ values for living
with the status quo and so make the gamble of war more attractive. More broadly, an important
feature of the baseline model is that it puts the guns-butter tradeoff at the heart of the theory, a
political economy dimension entirely missing from the “anarchy arguments” glossed above.12
Third, some implications of the model are new – such as the results on state size and arms
burdens – or revise existing theories. For example, existing conceptualizations of the offense-
defense balance define it in terms of the prospects or costs of successful attack versus defense,
taking force levels as exogenously given. But as becomes clear in the model, states choose force
levels endogenously in light of military technology and geography in order to affect prospects for
successful offense or defense. This point undermines or even inverts standard offense-defense hy-
11Jervis 1978; Glaser 1992, 2010; Kydd 1997, 2005.12There is no consensus formalization of the security dilemma and informal versions vary in their logics.
The standard brief definition, that it refers to the fact that “many of the means by which a state tries toincrease its security decrease the security of others” (Jervis, 1978, 169), is by itself a statement of a tradeoff
at best, not a social dilemma. The next step is usually to suggest that the tradeoff plus uncertainty about otherstates’ types leads to unwanted, dangerous mutual escalation. But it is not clear why arming due to mutualfears of expansionist types should increase fears further (a “spiral”) unless we assume a particular formof irrationality or suppose that states are also uncertain about others’ beliefs about their own types (Kydd,1997). Even in the latter case, escalation correctly signals a higher chance that the other is expansionist, andstatus quo states can have an incentive to signal their type by restraint (Kydd, 2005). Finally, in the Jervismodel tragic outcomes can occur only when both states are in truth status quo oriented; in the baselinemodel presented here, inefficient outcomes arise when one or both states have some revisionist preferencesand this is commonly known. As noted below, this may be a better fit for important empirical settings, suchas current US relations with China, Iran, or North Korea, for example.
6
potheses about how these factors matter. For instance, relative offensive advantage in technology
or geography causes states to arm to higher levels, and at these levels the odds of successful attack
can actually be lower than if defensive advantages in technology were greater. Thus, railroads
increased rapid mobilization capacity in late 19th century Europe, which drove increases in equi-
librium armament levels, which in turn made the status quo less desirable but conquest less easy
than it would have been at lower arms levels.
Fourth, compared to existing theory, the baseline model yields more determinate and testable
implications for a measurable proxy for the costs of anarchy – military burdens and force sizes.
For lack of space I only briefly consider broad patterns in military spending and force size among
the major powers over the last 150 years in light of the model’s implications, leaving an assessment
based on panel data to other work.13
Regarding the model, the closest papers in the literature are due to Powell and Matthew Jack-
son and Massimo Morelli.14 Powell’s underappreciated article was the first to explicitly model
state choices about how much to arm and whether to attack in a fully dynamic setting. He iden-
tified the tradeoff that lies behind the war constraint in the model below: higher peacetime arms
13For a panel data analysis see the draft version of this paper, Fearon 2015.14Powell 1993; Jackson and Morelli 2009; see also Morrow 1997. Also related are contest models such
as Skaperdas 1992, Hirshleifer 1995, and Garfinkel and Skaperdas 2007, where decisions about arms orpredation effort translate directly into resource allocations. As here, bargaining conditional on arms levelsis assumed to be efficient, but there is no possibility of “arming up” and attacking. The model below canbe seen as a repeated contest model with the added possibility of attack, which implies a war constrainton minimum arms levels. Note also that arms do not accumulate across periods in any of these models(including here), so they cannot easily be interpreted as models of arms races. When dynamic, they areinstead about repeated decisions on military spending rather than military stocks. Richardson arms racemodels focused on accumulation of arms stocks but did not model any decisions, including attack; see Isardand Anderton 1985 for a review. Another literature analogized arms races to repeated Prisoners’ Dilemmas,representing neither accumulation nor decisions to attack; see for examples Axelrod 1984; Downs and Rocke1990. Fearon 2010 analyses a game-theoretic arming model with both accumulation and explicit attackdecisions. Slantchev 2011 studies arming in crises as a signaling strategy, and, in a model with one roundof arming, Kydd 2000 considers bargaining in the shadow of a possible arms race where there is uncertaintyabout one side’s capacity.
7
levels reduce an adversary’s ability to “break out” and attack, but they also reduce the state’s values
for living with the status quo by reducing consumption. The costs of deterrence can make war to
eliminate or lessen the need for deterrence a more attractive, if tragically inefficient, option.15
Whereas in Powell’s model the only issue at stake is control of one’s own or the other state’s
territory, the model here adds an international issue over which the states have conflicting prefer-
ences. This could be anything, for example, the terms of a trade deal, the division of Poland or
spheres of influence, support for rebels in a third country, or the value of getting one’s way on all
disputed issues apart from control of homeland territory. This addition allows the model to speak
to the IR debate on cooperation and conflict as it developed after Waltz. Arming in the model is not
just about attack or deterrence, but also about gaining bargaining leverage on contested issues. The
states have an incentive to arm for bargaining leverage, but wish to avoid wasted effort. In effect
they face a classic “repeated PD” problem with respect to the contested issues, but this problem is
embedded in a larger context where the use of force puts sovereignty at risk.
Jackson and Morelli examine models similar to Powell’s but their states choose arms levels
simultaneously rather than sequentially. The minor timing difference proves to have oddly large
consequences. With simultaneous arming there is no pure strategy equilibrium because at levels
high enough that neither wants to break out and attack, a state could do better by temporarily
lowering arms spending before the other has a chance to react. This may seem like a technical
artifact, but it raises a good substantive question: If arms levels are just about deterrence, why
aren’t states constantly trying to save on spending by short-run cuts not large enough to draw
an attack? The model below suggests one possible resolution. Arms are valuable not only for
15See Coe 2012 on “costly peace” arguments for conflict more generally, and Kydd 2015, chap. 7, whoargues for the term “costly deterrence.” The core mechanism was characterized by Kant in Perpetual Peace:“For they [states with standing armies] constantly threaten other states with war by the very fact that they arealways prepared for it. They spur on the states to outdo one another in arming unlimited numbers of soldiers,and since the resultant costs eventually make peace more oppressive than war, the armies are themselves thecause of wars of aggression, which set out to end burdensome military expenditure.”
8
deterring but also for bargaining leverage. Cuts that would not lead to an all-out attack by a rival
would nonetheless disadvantage the state in issue bargaining. The model has simultaneous arming
decisions, but stable arms levels when peace is supportable.16
The literature on “the bargaining model of war” focuses on frictions that might explain costly
armed conflict, such as private information and incentives to misrepresent, and commitment prob-
lems occasioned by power shifts.17 While war can be an equilibrium outcome in the model pre-
sented below (arising from “costly deterrence,” a commitment problem) its focus is not on explain-
ing specific wars but instead is about the quality of interstate relations more generally. The main
choice of interest is arms levels (or military effort), which emerge as a theoretically defensible
measure of the degree of inefficiency in an interstate system that might be attributed to, or blamed
on, the condition of anarchy. The model highlights how arms levels feed back on states’ values for
living with a peaceful status quo and thus affect their willingness to run war risk. But specific paths
to war are not the focus (as indicated, for example, by the assumption of efficient issue bargaining).
The next two sections define the model and give equilibrium results. I then present com-
parative statics on what determines maximum feasible interstate cooperation, compare the logic to
existing IR theory, and provide a brief assessment of how the comparative statics square with broad
temporal and cross-sectional empirical patterns. The fifth section summarizes results for a variant
in which states differ in relative resources, and briefly discusses arguments about multipolarity.
16Meirowitz and Sartori 2008 and Debs and Monteiro 2014 show that when arming decisions are not ob-servable, pure strategy equilibria may disappear and war may occur even when arms are useful for bargainingadvantage.
17For examples, Fearon 1995; Reiter 2003; Powell 2006. Models in this literature assume, realistically, aknown conflict of interest, similar to the model analyzed here and distinct from standard security dilemmaarguments.
9
Model
Two states, 1 and 2, interact in successive periods t = 0, 1, 2, . . .. They each get resources normal-
ized to size 1 each period. Resources can be consumed or converted, one-for-one, into arms. For
simplicity and clarity, in this section the two states are identical in terms of resources and other
parameters, such as value for taking territory.
In each period, the states simultaneously choose arms levels ati ∈ [0, 1], next observe at =
(at1, a
t2), and then simultaneously choose whether to attack. (When there is no need to be explicit
about the period, I will use (a1, a2) for arms levels.) If either attacks, war occurs, one side wins,
taking all the resources or territory, and the strategic interaction ends. If neither attacks they move
to the next period. Payoffs are discounted by the common factor δ ∈ [0, 1) each period.
In every period t, there is an international issue worth γ > 0 that can be divided up between
them. This could stand for a different issue in every period, or it could be the same issue, always
subject to renegotiation. The most natural interpretation is that γ stands for many issues, summa-
rizing their total value to the two states. In the discussion of empirical implications below, I will
refer to γ as the gains from trade. Conceptually, the distinction between γ and the resources that
each state controls is that the states are “sovereign” over the latter, but cannot unilaterally decide
the allocation of γ.
Since the focus here is not on inefficiencies in bargaining over γ, I assume that if war does
not occur, the states divide up γ in each period as a function of relative military strength. Suppose
that if force levels are (a1, a2) and there is no war, then state 1’s payoff is 1 − a1 + γq(a1, a2) while
state 2 gets 1 − a2 + γ(1 − q(a1, a2)). This says that a state’s payoff if war does not occur is what
it doesn’t spend on arms plus its value for the issue resolution. A number between zero and one,
q(a1, a2) is state 1’s share of the division of the issues γ.
10
q(a1, a2) has the usual properties of a “contest success function.”18 It is symmetric, which
implies an equal split when a1 = a2, and q(a1, a2) is increasing in a1 and decreasing in a2. Thus a
state can gain a larger share on the issues by increasing its arms. q(a1, a2) can be interpreted as the
probability that state 1 wins the (limited) military conflict over γ that would occur if bargaining
failed, and thus a component of the “disagreement point” that determines the bargaining outcome.
If state i ∈ {1, 2} chooses to attack j in a period (and j stays on defense), then we will assume
that i wins the all-out war that occurs with probability pi(ai, a j; m), where m ≥ 0 is an exogenous
parameter that indexes offensive advantage. A factor will be said to increase offensive advantage if
for any given positive force levels at the time of attack, its presence improves an attacker’s odds.19
To give a specific example, consider
pi(ai, a j) =ai
ai + a j/m. (1)
Thus for values of m less than 1, the defending state j gets an advantage against the attacking state
i, for (say) equal force sizes. In the specific case of (1), m can also be interpreted as a measure of
“the offense-defense balance” in that 1/m is the ratio of attacker to defender forces that gives an
attacker a 50-50 chance of victory.
Finally, state i’s expected payoff from period t forward if it attacks defender j in period t is
1 − ai + pi(ai, a j)γ − c + δ(1 + µ)
1 − δ.
18Hirshleifer 1989.19Formally, m′ > m implies pi(a1, a2; m′) > pi(a1, a2; m) for all (a1, a2) > 0. See the online appendix for
further technical conditions on pi used in some propositions below. A second type of “offensive advantage”pertains to factors that increase the mobilization increment an attacker can get before counter-mobilizationby the defender. We could set ∆ > 0 as the largest amount a state can add to its forces in a single period (ifit has the resources). For simplicity I will allow states to arm all the way to ai = 1 in any period. Results forthe model with ∆ as a constraint representing this aspect of offensive advantage are the same as for m.
11
For simplicity war is treated as a costly lottery. The loser is eliminated, getting a payoff of zero
henceforth. The winner gets full control of the issues in the present and all future periods, as
well as the loser’s territory and resources starting in the next period.20 We assume that each state
values the other side’s territory at µ ∈ [0, 1], which allows us to vary how “greedy” they are.21
Small µ means that they don’t get as much value out of ruling the other state’s territory as they do
their own (which is worth 1). This could be for several reasons, including costs of ruling hostile
foreigners, lack of interest in territory not considered to be historic homeland, or democratic values
that support self-determination or political equality. The parameter c ≥ 0 is broken out to represent
additional costs of fighting, such as destructive impact.22
I focus on conditions for an undominated pure-strategy subgame perfect equilibria with no
war, and conditions in which war is the only equilibrium outcome.23
By assuming that the most a state can arm in a period is 1, we have also assumed that whatever
value the states derive from the issue dimension, it cannot be converted into military capability.
This is not necessary. We could have ati ∈ [0, 1 + γq(at−1
i , at−1j )], so that last period’s total resources
can be used for this period’s arms. Allowing γ to be usable for military capability may or may not
tighten the war constraint in the analysis that follows, but does not change qualitative results.
20Alternatively we could assume that the victor replaces the loser’s regime with a new government thathas different territorial or issue preferences; this model does not address the post-war choice. See Bueno deMesquita, Morrow, Siverson et al. 2003, chap. 9.
21Glaser 2010.22War costs can be introduced in several ways. They could be paid by both sides (here only winner pays,
while loser is just eliminated), or we could reduce all values of the winner by a factor. In the symmetriccase it doesn’t matter. The assumption that costs are effectively paid forever is for convenience. Below wedo the analysis taking the discount factor to 1, which means that any cost paid in a finite number of periodseffectively vanishes. Permanent damage is simple way to keep a cost term in the easier-to-present modelwith δ→ 1.
23Because the states simultaneously choose whether to attack, there are always equilibria in which bothattack because they expect the other to attack, even though neither attacking is part of an equilibrium thatboth prefer.
12
For simplicity and generality the bargaining over the issues represented by γ is treated in a
reduced-form way. As discussed in the online appendix, several common bargaining protocols can
give rise to a function q(a1, a2) as described. These include a stage game in which each state has
an equal chance of finding itself in a position to change the status quo on the issue and the other
can then choose whether to use force to try to reverse a fait accompli; and a game where we simply
select the Nash bargaining solution over division of γ, with disagreement being a costly conflict
lottery over γ. The key assumption, which is entirely consistent with the Realist tradition, is that
relative military power translates into more favorable issue outcomes on average. This is not true
for all issues all the time; for instance, if both sides must agree to change the status quo, then
a challenger’s threats to use force may be hard to make credible. The assumption here is that it
is true often enough that states expect that reducing military strength could disadvantage them in
issue bargaining in the future.
Results
How much “cooperation under anarchy” is possible in this two-state model? Note that if peace can
be sustained (no attacks in equilibrium), then cooperation can be identified with the equilibrium
level of arms. More arms, less cooperation. This is because arms spending is just wasteful. If the
states spend positive amounts (a1, a2) and divide γ in shares q(a1, a2) and 1 − q(a1, a2), they could
be better off spending at lower levels that yield the same division. The most efficient outcome, or
maximum cooperation, would be a peaceful equilibrium in which neither spends anything on arms
and the issue is divided equally in every period, yielding per period payoffs of 1 + γ/2.
Two temptations work against feasibility of the first best. The first is the war constraint: if
one state has no arms then the other may prefer to build and attack, so as to gain control of the
issues and more territory (if it values the latter). The second temptation is that independent of the
13
incentive to arm and attack, arming would confer bargaining leverage and yield more favorable
issue resolutions.
Cooperating on international issues
It is useful to begin by forgetting the war constraint and considering the problem of coop-
eration over the issues. A state can get more favorable issue resolutions by increasing military
strength, but if the other replies in kind both are worse off than if they had not armed up. If they
fail to construct what would amount to an arms control agreement, a spheres of influence agree-
ment, an agreement to demarcate or demilitarize a border, or the like, then the non-cooperative
outcome is the pair of arms levels such that neither wants to spend more given what the other
is spending. This is the level aNE (“NE” for Nash equilibrium in the stage game) such that the
marginal issue benefit of spending more equals the marginal arms cost of 1 when both choose it
(assuming an interior solution):
γq1(aNE, aNE) = 1.
If, as is natural, we assume that the bargaining returns to a given increment of arms are lower at
higher levels, then it is easy to show that the more the states care about the issue conflict (larger γ),
the more effort they waste on the competition (larger aNE).24
However, as Axelrod, Keohane, and others argued, the states can use “the shadow of the
future” to construct deals to support more cooperative outcomes. They agree to spend or compete
less and to compromise on the issues, with the understanding that if either reneges they will return
to costly competition. Consider an arms level a < aNE. Using a trigger strategy as the implicit
punishment threat, the states prefer to stick with cooperating at a provided that
24The assumption is that q1(a, a) is decreasing in a. See the online Appendix for details. Subscripts onfunctions indicate derivatives (qi(ai, a j) = ∂q/∂ai).
14
1 − a + γ/21 − δ
≥ maxa
1 − a + γq(a, a) + δ1 − aNE + γ/2
1 − δ.
The left-hand side is the payoff for sticking with the deal. The right-hand side is the best a state
can do by arming up in a period, which will then be followed by a return to costly competition.
Multiplying through by 1 − δ and considering what happens as δ approaches 1 – that is, as states
put more weight on their welfare in future periods, or as the time it takes to perceive and respond
to arms changes by the other side lessens – we get the following result.
Proposition 1. If military power can be used to influence issue resolutions beyond state borders
but not for conquest of the other state’s territory, then for large enough δ the states in the model
can support cooperation on any a < aNE, including maximum cooperation at a = 0 in all periods,
as an equilibrium outcome.
So far, so neoliberal. If the “if” condition in the proposition held universally, the fact of
anarchy would in principle pose little or no problem for states in international politics, and there
would be no grounds for viewing international relations as a necessarily tragic realm.
But that condition need not hold. If state j has no military, state i might prefer to arm and
attack to take territory. To keep things simple, suppose that pi(ai, a j) is such that when a j = 0 state
i can win for sure at minimal cost in arms.25 Then i prefers to arm a little and take over j if
1 + γ/21 − δ
< 1 +γ − c + δ(1 + µ)
1 − δ,
which is the case if
c < δµ + γ/2. (2)
25That is, pi(ai, 0) = 1 for ai > 0. This isn’t strictly necessary; we could have that i may require somepositive arms level to take over a j with no formal army.
15
This condition means that if the costs of fighting do not outweigh the value the states put on
acquiring new territory and full determination of the issues, the first-best outcome of maximal
cooperation (no arms spending) and no war is not possible, because each would be tempted to arm
up and invade.
The war constraint
From the analysis above it is evident that given (2), stable peace requires that both states arm
enough that neither would want to “break out,” arming rapidly and attacking, as Hitler did in the
1930s. Putting the incentive to compete on the issues to the side for the moment, suppose both
states are choosing arms level a > 0. For this to be an equilibrium each must prefer peace at a to
mobilizing and attacking:
1 − a + γ/21 − δ
≥ maxa
1 − a + pi(a, a)γ − c + δ(1 + µ)
1 − δ. (3)
This is the war constraint, the condition on arms that must be satisfied for states not to want to
mobilize and attack. Multiplying through by 1 − δ, we have
1 − a + γ/2 ≥ maxa
(1 − a)(1 − δ) + pi(a, a)(γ − c + δ(1 + µ)). (4)
Although not necessary for the results that follow, we can simplify and get a clearer under-
standing of the strategic dilemma by considering the problem for discount factors close to one. If
the “shadow of the future” is long enough, a state maximizes the value of the break-out option by
spending all available resources on arms in a period (i.e., a = 1) in order to maximize the chance
16
of victory in war.26 So the war constraint simplifies to
1 − a + γ/2 ≥ pi(1, a)(γ − c + 1 + µ). (5)
Figure 1 graphically depicts the war constraint in two possible cases. In the first, there are
values of a that satisfy the constraint and there is a minimum value a∗ that satisfies it with equality.
In the second, there is no arms level high enough that neither side prefers to mobilize and attack
but low enough that a peaceful status quo is preferable to trying war in hopes of getting rid of the
need for costly deterrence.27 In this second case, war is inevitable; in fact, it must occur in the first
period in equilibrium. The cause can be attributed to anarchy (and condition 2), in that the states
would prefer it if they could commit to zero arms and no attacks. Note also that the “shadow of
the future” is as long as it can be here. For preferences and a military technology such that the war
constraint cannot be met, the states’ capacity to attack and possibly eliminate the other undermines
the possibility of cooperation based on repeated interactions.28
Suppose parameters are such that the war constraint can be satisfied. This is necessary for a
peaceful equilibrium to exist, but not quite sufficient: It must also be that neither state would want
to decrease its arms level enough in one period to save on costs but not so much as to draw an
attack. As the following main result details, neither wants to do this if reducing arms spending
26The marginal cost of arming on the right-hand side of (4) is 1 − δ, which gets arbitrarily small as δ goesto one. So as long as pi increases in its first argument, the a that solves the RHS of (4) must be a = 1.A better formulation (which I don’t use here for simplicity) would have that states can add no more thana fixed amount ∆ > 0 in one period (before the other can effectively counter), in which case ∆ measures asecond form of “offensive advantage.” This form is referred to below in the discussion of railroads and othertechnological aids to mass mobilization in the late 19th century.
27This is the tradeoff that Powell 1993 identified. Due to the timing difference (simultaneous armingrather than sequential), satisfying (4) is not sufficient for equilibrium here, though it is in Powell’s model;see below.
28Powell 1993.
17
would hurt the state enough in bargaining over disputed issues.
Proposition 2. If there is a smallest a∗ that solves the war constraint (4) with equality, and a∗ ≤
aNE, then for large enough δ the game has peaceful equilibria in which, on the path of play, the
states choose a ∈ [a∗, a] in every period, where a ≤ aNE is determined by specific parameters.
Thus in the most cooperative equilibrium the states choose a∗ in all periods on the path. Further,
if defensive advantage is sufficiently large (m is small enough), such an a∗ definitely exists and
a∗ < aNE.
Using standard assumptions about states’ preferences and capabilities, Proposition 2 provides
an answer to the question of how much cooperation under anarchy is possible. As neoliberal insti-
tutionalists argued, states can use repeated interactions and the shadow of the future to construct
regimes or tacit deals that make them better off than they would be without such formal or infor-
mal institutions. (By “institutions” I mean simply coordination on more efficient outcomes that
requires acting against short-run interests; in practice this is often facilitated by international orga-
nizations whose roles are not modeled here.) Without institutions, states in the model engage in
costly competition at the myopic Nash outcome of aNE.
However even with institutions – coordination on an a∗ < aNE – there are still unavoidable
costs of anarchy. The depth of cooperation is constrained by the inability to commit not to go to
war if the other state’s military effort is low enough.
Before considering what determines how deep interstate cooperation can be, a comment on
the strategic role played by issue competition. If there are no disagreements over international
issues (γ = 0), then the cooperative equilibria just described do not exist. With γ = 0 the states
have no reason to arm for bargaining leverage (aNE = 0), so there is no immediate penalty for
short-run cuts not large enough to undermine deterrence of invasion. For small enough γ but
enough “greed” about territory that condition (2) still holds, the model becomes close to that of
18
Jackson and Morelli.29 Mixed strategy equilibria result, with positive probability of attack and
war. Empirically, however, we do not see states constantly and arbitrarily varying their military
spending.30
Introducing issue competition provides one empirically plausible way to resolve the puzzle.
When a∗ < aNE, undercutting on military strength leads to less favorable outcomes on international
issues the state cares about. In other words, arming or other costly competitive effort is valuable
not only for deterrence of invasion but also for helping to get one’s way in the random and di-
verse disputes that make up international affairs. “Diplomacy without arms is like music without
instruments,” as Frederick the Great put it.31
In this deliberately spare, “pure IR” model, a state’s only reasons to want arms and a military
have to do with foreign policy. They are used to deter or attack another state and to gain leverage
in international bargaining. But of course rulers want armies for domestic political reasons as well.
They need guns and soldiers to deter or put down organized rebellions. For many countries, do-
mestic demand for arms could be the explanation for why they don’t have an incentive to undercut
to save money, getting into the unstable dynamic discussed by Jackson and Morelli. Let aD be the
spending level the state’s leadership would choose if there were no international threats or bargain-
ing considerations whatsoever. If a∗ < aD, then the same argument as given above would apply,
although in this instance international cooperation would be of no use for reducing arms levels to
the minimum that satisfies the war constraint. To the contrary, such a domestically insecure state’s
demand for a larger-than-necessary military might aggravate international conflict by increasing
29Jackson and Morelli 2009.30Kydd 2015, 121, makes this same point.31If a∗ > aNE the game does not have a pure strategy equilibrium and I will not attempt to characterize the
mixed strategy equilibrium that results; see Jackson and Morelli 2009 for an analysis of a simplified versionwith three arms levels. Another approach to resolving the mixed-strategy puzzle is to introduce uncertaintyabout state types, so that if a state deviates down it increases the probability that the other will be a type thatwould then want to attack. See Kydd 2015, 120-124.
19
neighbors’ military burdens.
As discussed further below, in the post-Cold War world many small states’ borders are de
facto protected by the expectation that the U.S., the European Union, and perhaps the U.N. would
impose sanctions or worse if they were openly invaded by a neighbor. (In the two-state model, this
amounts to aggressors having higher costs to attack due to expected reactions of more powerful
states.) These states’ arms levels may be entirely determined by domestic political considerations
and not at all by considerations of deterrence or competition on interstate issues.
How deep can interstate cooperation be?
Focused on the claim that anarchy makes international politics inevitably conflictual, Waltz had
little to say about what explains variation in international cooperation. His main variable was the
number of major powers (more on which below). In arguments that greatly influenced Realists, he
suggested that “unit level” factors – characteristics of states other than relative power, like regime
type or territorial preferences – should not be part of a “systemic” IR theory and also are not likely
to have large and consistent effects on levels of conflict and cooperation.
Writing at about the same time, Jervis argued that variation in the extent of international
conflict and cooperation depends on factors influencing states’ ability to arm up and invade, and
also to distinguish offensive from defensive forces. For example, states separated by distance and
water should find it easier to cooperate, as should nuclear weapons states with secure second-strike
forces, because these make invasion and conquest nearly impossible. Jervis’ logic relied on the
idea that states are always uncertain about whether their counterparts might be aggressive types.
When offense is advantaged, it is more dangerous to cooperate (e.g., to lower your arms levels)
because you are in deeper trouble if the other side turns out to be an aggressive type.32
32Jervis 1976, 1978. The “uncertainty about adversary intentions” version of this argument has been
20
The main variables determining the feasibility of international cooperation for neoliberal insti-
tutionalists were the “shadow of the future” (state leaders’ discount rates) and a somewhat generic
notion of degree of conflict of interest.33 Whether because the shadow of the future is difficult to
measure or because leader discount rates don’t explain much variation in international cooperation,
the cooperation-under-anarchy program has been more successful at explaining the basis for much
international cooperation and many international institutions than at explaining variation over time
or across states.
What determines maximum feasible cooperation in the model presented above?
Proposition 3. Maximum feasible international cooperation (smaller a∗) is (1) decreasing in the
value the states derive from controlling the other’s territory (µ); (2) increasing in defensive advan-
tage (smaller m) and war costs (larger c); and (3) may increase or decrease with the value of the
international issues in dispute (or the gains from trade) between the two states, γ.
Value for territory
The less “greedy” the states are about additional territory, the less they need to arm for de-
terrence (or otherwise compete) and thus the greater the scope for cooperation on other issues.
This is an intuitive implication of this fundamentally Realist model, and it is consistent with the
claims of neoclassical realists who argue that most international conflict stems from states with
revisionist preferences rather than “tragic” conflict between fearful, uncertain security-seekers.34
In the model, however, it is precisely revisionist preferences that make for tragedy, in the sense
developed most extensively by Kydd 1997, 2005, Glaser 1992, 2010, and Baliga and Sjostrom 2004, 2011.Jervis’ arguments can also be interpreted in terms of the equilibrium selection principle of “risk dominance”;see for example Padro i Miquel and Chassang 2007, 2010, who get to risk dominant equilibria via a global-games approach.
33Oye 1986.34Schweller 1996, 1998; see also Kydd 2005.
21
of inefficient arming that makes war more attractive and that could be avoided but for interstate
anarchy. Here, more “greed” makes for more inefficiency, not less. (Note that standard security
dilemma arguments conceptualize inefficient outcomes only in terms of conflict between security
seekers.)
The mechanism is not simply that more value for additional territory makes conquest more
valuable, but instead involves a feedback loop. Suppose a new leadership comes to power in state
A that puts greater intrinsic value on controlling territory beyond its current borders (an increase
in µA in an asymmetric version of the model). At prior arms levels, state A would want to arm
up and attack. Understanding this, state B arms to a higher level to deter A. But this makes a
peaceful status quo less attractive for state B since it now has to spend on an on-going policy of
“containment.” So in turn A needs to arm more to deter attack by state B, even if state B is a pure
status quo type that has zero intrinsic value for controlling more territory (µB = 0). In turn, A’s
need to deter B from attacking to get rid of the threat arising from A’s revisionist preference lowers
A’s value for a peaceful status quo relative to war, feeding back again on B’s need for arms (and so
on up to equilibrium levels, if a peaceful equilibrium exists).
Two points follow on this example. First, it shows how the model captures the Waltzian
insight that “systemic forces” can override the influence of particular “unit level” state desires
or characteristics. A state with no intrinsic value for acquiring more territory or deposing another
regime might be driven to do so by a neighbor’s greed along with system effects of its own response.
The idea of “liberal imperialism” or “making the world safe for democracy” follows this logic.35
Second, it nonetheless does not follow that the intrinsic value states put on territory is irrelevant for
the level of international cooperation, or that all states are driven by anarchy to be expansionist, as
35A developed example is Coe’s account of the reasons for the U.S. invasion of Iraq in 2003. He arguesthat the Bush administration judged (poorly, it would seem) the costs of continued containment of SaddamHussein’s regime to be greater than the costs of deposing him. The costs of containment stemmed from thesame sort of commitment problem as in the model above. See Coe 2012.
22
“offensive Realists” like Mearsheimer maintain.36 Rather, this state-level characteristic is a natural
component of a Realist, structural model, and its variation has an important influence on outcomes.
There has been little analysis in IR of the determinants of states’ intrinsic value for control-
ling additional territory, whether by critics of Realism or Realist critics of Waltz’s and offensive
realists’ skepticism. And perhaps due to lack of theory linking revisionist territorial preferences
to the costs of anarchy, we have not seen extensive efforts to measure variation in territorial pref-
erences to check whether this might explain significant variation in the costs of anarchy.37 The
baseline model suggests how the international effects of plausibly important state-level factors –
such as democracy, nationalism, and industrialization – can be incorporated in a structural model
and analyzed for their impact on the costs of anarchy, via their effect on states’ values for territorial
conquest. I provide a very partial sketch in the rest of this subsection.
What determines a state’s value for ruling new territory? In ancien regime days it might have
been the productivity of the peasants and towns that a ruling noble could tax. Nowadays it can still
have to do with the material value of land. For instance, high arms levels around the Persian Gulf
are plausibly related to the presence of valuable oil reserves. But probably more often it has to do
with nationalist ideologies and attachments. Where there is agreement that borders correctly di-
vide national territories, values for adding more are lower than where there is disagreement, other
things equal. Empirically, arms levels tend to be much higher while cooperation and international
institutions are less evident where there are dissatified nationalisms. Consider Israel and its neigh-
bors, North and South Korea, Eritrea and Ethiopia, Armenia and Azerbaijan, or India and Pakistan
(who disagree about Kashmir).
Table 1 lists the top and bottom 15 states on two measures of military effort – military spend-
ing as a share of GDP and armed forces per 1,000 population – for both the Cold War years and the
36Mearsheimer 2001.37There is not much in the IR literature besides Kaysen 1990 and Rosecrance 1987, 1999.
23
period 1991-2010.38 The first superscript on the country name indicates the number of territorial ri-
valries the state was involved in according to the codings of William Thompson and David Dreyer.
These are cases where Thompson and Dreyer judge that two states view each other as “threat-
ening competitors” and where they “contest the exclusive control of territory.” Most often these
territorial rivalries are linked to competing nationalist ideologies concerning what land properly
belongs to the nation, or in the case of the two Koreas (which Thompson and Dreyer do not code
as a territorial rivalry) which regime should rule the territory that belongs to the Korean nation.
Territorial rivalries are clearly much more common among the highly armed states than among the
low-military-effort states. Further, note that high military effort states are often the smaller state in
a dyad with an intense nationalist conflict – for instance, Eritrea, Armenia, Israel (versus its several
larger neighbors), Kuwait, Greece, Taiwan, and North Korea.39
Regarding variation over time, there are good reasons to think that industrialization, the spread
of democracy, and the sorting of people and boundaries according to nationalist ideologies have –
on average and over the last 200 years – reduced states’ intrinsic value for acquiring new territory.
Concerning industrialization, when income comes mainly from human capital and manufactured
tradables rather than land, it is arguably harder to profit from military conquest, and less necessary
since trade can provide a cheaper alternative.40 Next, democracies have less economic rationale for
38Military burden is the Correlates of War (COW) estimate of military spending divided by total GDP(World Bank), both in current U.S. dollars. I extend the series for 2008-2010 using SIPRI estimates. Overallcoverage is 1960-2010. Soldiers per 1000 comes from COW and is available for 1945-2007. Military burdenis missing for at least one likely top spender, North Korea. Finally, the list is based on countries that hada population greater than 500,000 in 1990 (or first year of independence if later), so 15 is approximately adecile.
39Thompson and Dreyer 2012, 21. By saying that a territorial conflict is nationalist I do not rule out thatthere may be material motivations as well, as in the case Iraq’s claim on Kuwait. Of course, the bivariaterelationship between rivalry and arms spending is very strong. Post-1945 arms levels are analyzed using themodel to guide the empirical specifications in Fearon 2015.
40On the impact of industrialization on the value of conquest, see especially Kaysen 1990, Gat 2005,Gartzke 2007, Coe 2015, and Liberman 1998 for a critique. In work on democratization Boix 2003 and
24
military conquest than do narrowly held autocracies if the rents must be shared more broadly, or if
a strong norm of political equality (which might have an instrumental basis) implies that people in
conquered territory would have to be treated as citizens on equal terms.41 Concerning nationalism,
the aftermath of two world wars, decolonization, and the break up of the Soviet Union all con-
tributed to greater congruence between interstate borders and conceptions of common nationality,
albeit with a number of important exceptions that continue to stimulate conflict and arming, as
noted above.
Figure 2 shows striking declines over the last 190 years in the number of territorial rivalries per
independent state, again using Thompson and Dreyer’s codings. The change is particularly large
for the militarily strongest states in the international system, which saw a steep and persistent drop
in territorial rivalries in the years around World War II. To be sure, this is not a perfect measure
of average values for µ, for reasons discussed below. But since territorial rivalries involve public
claims on a neighbor’s territory sufficient to cause perception of serious threat, their decline is at
least consistent with the proposition that on average states’ value for acquiring new territory has
declined. Further, the cases behind the large drop for major powers from 1939 to 1945 are consis-
tent with the idea that democratization and more nationalism-congruent borders have contributed
to the long-run reduction. These include the end of three disputes associated with Italy’s switch
from fascism to democracy; four territorial rivalries associated with the end of imperial Japan; and
two rivalries associated with end of Nazi Germany and the redrawing of borders (and possibly the
ethnic cleansing of German speakers) in Eastern Europe.
The offense-defense balance
Acemoglu and Robinson 2006 argue similarly that industrialization increases the exit options for wealthproducers which in turn constrains predatory states.
41Bueno de Mesquita, Morrow, Siverson et al., 2003, 419; Fearon, 2012; Jackson and Morelli, 2007;Zacher, 2001.
25
In the model, deeper cooperation is possible when the odds of successful attack are lower
for any given force levels, which is how greater defensive advantage is defined here. Since Jervis
the canonical justification for this claim has been that states are uncertain about others’ preferences
over territory or issues (µ and γ in the model), and guessing wrong is more dangerous when offense
is relatively advantaged. Extensions of the core idea hold that offensive advantage makes “spirals
of hostility” and preemptive war more likely between what are in fact security seeking states.42
A weakness in these arguments is that it is not clear why genuinely security seeking states
that have no territorial or issue conflicts would be unable to credibly signal this to each other over
time.43 Another issue is that it is not empirically plausible that any state has ever been a pure
security seeker. Surely all states have things about the political world they would like to change if
they faced no external constraints. Perhaps this is just a matter of degree, so that the Jervis-Glaser-
Kydd “two type” model is still a good approximation for some cases, such as U.S. policymakers’
thinking about the Soviet Union during the Cold War. But there are many other cases where one or
both states correctly understand that the other has significant revisionist preferences and these are
at the root of the conflict. As noted above, standard security dilemma arguments either rule out or
have no way of conceptualizing tragic outcomes (inefficient arming, conflict) in these settings.44
In the baseline model the states have no uncertainty at all about each other’s preferences and
neither is a pure security seeker (if µ > 0 or γ > 0). Thus the standard security dilemma arguments
do not apply. Instead, relative advantage for offense increases arming and lowers cooperation
because more arms are needed to deter break out, and deterrence is needed because the states have
some revisionist preferences. Worse, offensive advantage has a perverse multiplier effect: The
42Jervis 1978; Glaser 1997, 2010; Kydd 1997; Van Evera 1984; Posen 1993.43Kydd 2005.44Nor is it clear if standard arguments go through if “status quo states” are in fact a little greedy when it
comes to issues. For example, defense dominance – such as a “nuclear shield” – might make it safer for aslightly revisionist state to pursue aggressive policies abroad.
26
more arms a state needs to deter attack by a potential adversary, the less desirable living with the
status quo becomes for that state. This makes war relatively more attractive, which means that the
other state has to spend more, which makes the status quo less good for it, and so on. As Powell
showed and as seen in Figure 1, the net result can be that there is no arms level high enough to
deter break out but low enough to make peace preferable to fighting.
Literally interpreted, the “no peaceful equilibrium” case in Figure 1 may be quite rare. But it
is plausible that a state’s value for living with the territorial status quo – and thus its willingness
to risk war – depends in part on how much it has to spend to maintain it. The model clarifies
how states’ value for (living with) the status quo is endogenous to the strategic dilemma posed by
anarchy, with the political economy tradeoff between guns and butter playing a central role.45
What determines variation in the offense-defense balance and are these things related to ob-
served patterns of international cooperation and conflict? Answering this question requires, first,
additional interpretation of “offensive/defensive advantage,” a difficult concept that is the source of
much debate. In almost all cases, scholars define offense-defense advantage or “balance” in terms
of the ease or cost of offensive operations, either given a defender’s forces or without specifying
the force levels at which cost is to be evaluated. In the most detailed analysis, Glaser and Kauf-
mann define the balance as “the ratio of the cost of the forces the attacker requires to take territory
to the cost of the forces the defender has deployed.”46
45To be clear, whenever µ or γ is greater than zero, states in the model are greedy types in the sense ofsecurity dilemma theory; they put positive value on acquiring more territory or changing issue outcomes fortheir own sake. The parameters µ and γ index how “greedy” they are in that sense. But their “value for thestatus quo” in the sense of the payoff stream from a peaceful status quo is endogenous and can vary, becauseit depends on how much consumption they can get in peace (which depends on military spending).
46Glaser and Kaufmann 1998, 50. Glaser and Kaufmann list other definitions, most of which take the sameapproach though without being clear about force levels. Representative would be Van Evera 2001, 118, whosays “In this book, ‘offense dominant’ means that conquest is fairly easy, ‘defense dominant’ means thatconquest is very difficult.” See Davis, Finel, Goddard et al. 1998-1999 and Lieber 2000 for debate on themeasuring the offense-defense balance and additional references.
27
But defenders choose forces in light of what they think is needed to deter, and defender force
size strongly affects an adversary’s costs or ability to arm up to a force ratio favorable for attack.
So if a technological change renders existing force sizes too small for comfortable deterrence –
say, railroads make the movement and supply of mass armies more feasible – then states may
arm to higher peacetime levels so as to offset what would have been good odds for an attacker
“breaking out” from the old force levels. The cost or ease of offense from the new equilibrium
force levels thus need not be any different from what it was before the technological change that
favored offense. Put differently, the underlying technological influences on offensive prospects
(summarized by the exogenous parameter m in the model) have a direct, positive effect on the odds
of successful attack for any given force levels. But this direct effect can be undone by the indirect
effect a larger m has via increased equilibrium arms levels.
Indeed, the baseline model yields the following general implication, which is surprising and
interesting in light of the existing literature.
Proposition 4. Consider the model with any contest success function pi(ai, a j; m), where m indexes
offensive advantage in the sense that pi is increasing in m for positive arms levels. In a peaceful
equilibrium with arms levels a∗, the probability that a break-out attack would succeed is decreasing
in offensive advantage.
So greater offensive advantage implies higher equilibrium force levels (from Proposition 3),
but a smaller chance that an offensive would succeed at those force levels! The intuition is straight-
forward: Higher arms levels make the status quo less appealing. Thus for peace to continue to be
feasible at these levels, mobilizing and going to war must become less attractive, which means that
its odds of success must decline. The effect of the endogenous increase in arms levels must more
than offset the direct impact of an offense-favoring technological change.47
47To prove Proposition 4, write (5) as an identity using a∗(m) for a, and let p(m) ≡ p(1, a∗(m); m). Differ-
28
To illustrate, consider two examples, the first hypothetical. Two kingdoms are separated by a
broad river, and given the river each requires a standing army of only, say, 1,000 soldiers to deter
the other from “arming up” and invading with the maximum capacity of (say) 5,000 soldiers. A
massive earthquake upstream causes the river to run dry. Now 1,000 is not enough to deter break
out because a defender no longer has the advantage provided by the river. So both sides arm to
a new equilibrium level at, say, 3,000 soldiers, which is much costlier to maintain. Because war
is now more attractive than before in terms of the long-run cost savings from defeating the other
kingdom, arms levels have to be high enough that break out and attack is even less likely to succeed
than before in order to maintain deterrence. This is so despite the fact that for any fixed force levels,
the elimination of the river unambiguously favors offense.
Second, consider that many have argued that Europe on the eve of World War I was charac-
terized by a high level of defense dominance, in fact if not in leaders’ perceptions. The evidence
comes from the war itself, which showed that greatly improved firepower (better rifles and ma-
chines guns) had rendered taking defended terrain extremely difficult. At the same time, however,
railroads and better roads had made it possible for the continental powers to mobilize vastly larger
numbers of soldiers and deliver them to a front much more quickly than before the 1860s. This
meant that at low initial force levels, a break out – mobilization and rapid delivery of a mass army
to the front – could give a state a good chance of victory by a crushing blow, even despite some
machine guns on the other side. To deter, states thus needed to arm up to peace-time levels high
enough that such a course would not be sufficiently promising.
Thus the conventional IR wisdom about this famous case may have it backwards. A situation
of increased offensive advantage may have caused states to try to arm to levels high enough that
entiating both sides in m, p′(m) must be negative since from Proposition 3, da∗(m)/dm is positive. Note alsothat this result does not depend on the assumption that “break out” entails spending all resources on arms; italso obtains in the variant of the model wherein states are limited in the amount they can arm up before theother state can respond.
29
break out would not be so promising as to make attack appealing. There could have been increased
offensive advantage even though at equilibrium force levels driven by this technological condition,
offense was not highly likely to succeed.48 Arms levels determine the likelihood that an attack
will succeed, but at the same time the likelihood that attack will succeed determines arms levels.
In equilibrium, it may appear that “defense dominance” prevails, because many machine guns
and sophisticated use of railroads by a defender could probably stop an attack. But the need for
many machine guns and many quickly mobilizable troops in peacetime is driven by an offensive
advantage (the ability to generate a massive, mobile army quickly). Further, high levels of costly
arms and military preparedness, or the anticipation of ruinous arms racing, makes a peaceful status
quo less appealing.49
To summarize, offense is more advantaged the greater the odds that an attacker wins for any
given force levels, and the more forces mobilization can add before an adversary can effectively
counter. Greater offensive advantage in these two senses imply higher equilibrium arms levels
and other forms of costly competition (less international cooperation); probably greater odds of
large-scale war; but not greater odds that attackers will actually win their wars.50
With the offense-defense balance so understood, in a conventional world any technological
48Although some recent work by historians casts doubt on this view (see Lieber 2007), it could be that a“cult of the offensive” meant that leaders had radically incorrect beliefs about the prospects for offensivesgiven force levels and military technology in 1914 (Snyder 1984; Van Evera 1984). I am stylizing theexample (by assuming no cult of the offensive) to help make the point that offensive advantage in technology,geography, and so on, and the odds that an attack succeeds at equilibrium arms levels are different things.
49Drawing on recent work by historians, McDonald 2011 makes a compelling case that in the five yearsprior to July 1914 “the arms race on land had thus pushed Germany to its financial limits and threatened thegovernment’s capacity to sustain it,” inclining its leaders towards a preventive war with Russia, which wasarming up at a rapid clip and was not so financially constrained.
50In this complete-information model war either occurs or does not occur for given parameter values, andhigher m makes it more likely that no peaceful equilibrium exists. Empirically it is more plausible thatgreater offensive advantage increases war risk not by eliminating peaceful equilibria, but rather by makingstates willing to bargain more aggressively because they find deterrence more costly. See Fearon 2012 for aversion of this model with incomplete information bargaining that illustrates this mechanism.
30
or doctrinal innovations that enable states to mobilize and deploy troops more quickly favor of-
fense and so should make for higher arms levels and less international cooperation. By contrast,
because the threat to use nuclear weapons becomes increasingly credible as a nuclear-armed state
loses homeland territory in conventional war, nuclear weapons make old-style territorial conquest
difficult to imagine. As Jervis and others argued, this technological development implies a major
shift in the direction of defense dominance among nuclear states, and probably even among states
that could acquire nuclear weapons on short notice. With secure second-strike forces, a break-out
attack to overwhelm an adversary becomes suicidal for any disparity in conventional arms.
The empirical expectation, then, is that prior to the nuclear era the major powers should have
seen increasing arms levels and decreasing international cooperation after Prussia’s demonstration
of how railroads, good roads, and organization could be used to rapidly deploy mass armies in
the Austro-Prussian and Franco-Prussian wars of 1866 and 1871. Onorato, Scheve, and Stasavage
document that the share of major power populations mobilized during interstate wars increased
enormously in the period from 1859 to 1970 (or 1954), compared to wars in previous and subse-
quent years.51 They argue that this was largely due to the railroad. This is direct evidence for a
large increase in offensive advantage in a sense specified above – railroads and other technological
and organizational innovations caused a substantial increase in capacity to mobilize and deploy
large numbers of troops rapidly. Thus the results from the baseline model imply that we should
see, for the major powers, a general tendency for peacetime force levels to be increasing, until the
impact of nuclear weapons begins to be felt, after which there should be declines.52
The solid black lines in Figure 3 display armed forces per 1000 population for nine major
powers for 1816 to 2007, during years when the state was not engaged in an interstate war. The
51Onorato, Scheve, and Stasavage 2014.52Onorato, Scheve, and Stasavage link the decline of wartime force sizes to the development of precision-
guided munitions and cruise missiles in particular. Few of the several post-1971 interstate wars were foughtbetween adversaries fielding such weapons, however.
31
lines have wide dashes for years of interstate war. The small-dashed line is a smooth of the solid
black line, thus showing the trend for peacetime armed forces in the state (wartime force levels are
not included in the smooth).53
The patterns – mainly “inverse U” shaped – are largely though not entirely consistent with the
expectations based on the broad system-level variations in the offense-defense balance described
above. First, all of these states see declines (and mainly large declines) in armed forces from the
mid-1950s to the present. Although in a few cases the decline occurs mostly after the end of the
Cold War, for most it takes place throughout. This is consistent with the major powers adjusting
gradually to a new world in which, in contrast to the period from (at least) 1900 to 1950, they do not
need to worry about being able to quickly raise a mass army to defeat a possible attack by another
major power in a total war. Of course, other factors may contribute to more cooperation and less
competition among the major powers since World War II. More are democracies and there are
fewer nationalist disagreements about which nations should own what land in Europe or Northeast
Asia since the post-war settlements. As argued above, both factors may lower values for annexing
territory, which in the baseline model implies lower arms levels. But the pattern is nonetheless also
consistent with the large increase in defensive advantage due to the advent of nuclear weapons.54
On the other side of 1945, in various degrees France, Britain, Germany, the U.S., Japan, and
53To make the graph more readable I have coded the U.S. Civil War as “interstate” here. Otherwise, datais from the Correlates of War’s list of interstate wars and the National Military Capabilities data set, version4.0; Singer 1987.
54If nuclear weapons amount to nearly infinite defense dominance, why didn’t U.S. and Soviet forces fallfaster after the mid-1960s, and why haven’t more major powers acquired them? First, as argued above, statescan want arms for bargaining in disputes short of conquest or defense against conquest, and also for domesticpurposes. Second, U.S. Cold War strategy deliberately chose to maintain large conventional forces stationedin Europe and Asia, along with extended “security umbrellas,” rather than promote nuclear proliferation bymajor power allies like West Germany, Japan, and South Korea; see Trachtenberg 1999; Gavin 2012. Thatpolicy choice was the basis for large conventional forces in Europe during the Cold War, and, since 1991,what is called “unipolarity” (Wohlforth, 1999). Third, it has taken time for the major powers to adapt andhave some confidence in a nuclear world (as concerns major power relations).
32
Turkey (Ottoman Empire) all see increasing force levels in the years leading up to World War I,
consistent with the implications of increased offensive advantage in the second part of the 19th
century. Russia, Italy, and China do not, however, with these data. Russia may be explained by
the fact that its absolute army sizes were growing rapidly due to its large population and moderate
population growth (see the next section for results on relative size and arms levels).55 Note also
that after 1860 almost all European powers moved to conscription and shorter periods of service,
in order to increase reserve forces and the number of young men trained to fight. By contrast,
in the nuclear era the major powers have moved back to smaller professional armies with long
service periods, and smaller reserve forces.56 Historians have often stressed increasing military
competition and arms racing in the decades leading up to World War I.57
For the interwar period, the German army abided by the force size limitations of the Versailles
Treaty until 1933 when Hitler began his rearmament drive. Faced with a major power regime that
appeared to have greater value for territorial expansion than before, the other European powers and
the U.S. followed quickly, though not quickly enough. Joseph Maiolo details the politics of the
European and Asian arms races of the 1930s. Military and civilian leaders in all the major powers
had concluded from the Great War that “a modern war is a war of nations rather than armies,” so
that peacetime forces and military industry had to be capable of rapid expansion.58 He argues that
the strain of mobilizing to get to higher peacetime arms levels in the midst of the Depression was an
important factor driving Europe to war. The much larger mobilization capacity of a hostile Soviet
55The U.K. is demobilizing after the Crimean and Boer Wars, but appears to be on an upward trend evenso. In this era the U.S. and the U.K. had much smaller standing forces than the continental powers, whichis naturally explained in terms of the defensive advantage they get for being protected by oceans. Japansimilarly has relatively small peacetime forces as a share of population at all times. China does as well,though in this case a small population share translates to a very large army.
56Sheehan 2008.57See McDonald 2011 and references therein.58Maiolo 2010, 107.
33
Union meant that deterrence would require high and possibly unsustainable levels of peacetime
armament in Germany. Similarly, Dale Copeland shows how concern about the inability to match
Soviet military might in the long run motivated both Hitler and his generals to plan for preventive
war from 1933 on.59
Value for issues other than territory
The term γ represents the states’ value for issues other than homeland territory, and which
cannot be determined unilaterally by either state. The larger γ, the greater the potential for coop-
eration in the neoliberal institutionalist sense of agreements to share gains from mutual restraint
on competitive policies. The empirical example of greatest interest, given an extensive literature
and debate, concerns gains from international trade. Clearly, greater potential gains from trade
imply greater potential for international cooperation, as might occur if trade costs decline (due to,
for instance, steam power or containerized shipping or between contiguous as opposed to distant
countries). But how do greater gains from trade affect the costs of anarchy, here measured by a∗,
the smallest arms level consistent with a stable peace?
In contrast to value for territory or war costs, increasing the gains from trade has competing
effects. On the one hand, greater gains from trade increase the value of a peaceful status quo in
which these are shared, but on the other hand, there is more value to be captured from a successful
all-out war. The first effect favors lower arms levels, the second higher.
Virtually the entire literature on interdependence and war neglects the second effect. The
standard argument is that more trade between two states increases the opportunity cost of war, a
claim that does not consider what happens to trade after one side wins a war. If the winner takes
over or sets the policies of the loser, it can capture all gains from trade for itself.60
59Copeland 2000, chap. 5.60The most developed version of the opportunity-cost argument is Martin, Mayer, and Thoenig 2008. It
is perhaps more relevant for conflicts that do not put sovereignty at issue (which is true of most MIDs, for
34
The competing effects of greater gains from trade tip in the direction of relaxing the war
constraint the more the states have decreasing value for additional gains (the baseline assumes risk
neutrality for simplicity). This is because an international agreement that divides the gains gives a
state a “bird in the hand,” whereas trying to grab all via war is a gamble on “two in the bush.”
The results on γ also depend on how we envision domestic politics affecting the distribution
of gains from successful war. For an autocrat or a democracy willing to economically oppress the
people in a defeated country, successful war might double the “gains from trade” via the increased
tax base that results from conquest. But to the extent that the winning state would not dispropor-
tionately tax people in the defeated territory, then increased gains from trade simply increase the
value of a neoliberal deal between independent states, so reducing international competition (a∗).61
Finally, note that increasing the gains from trade in the model is not the same thing as reducing
protectionism. The above analysis assumed that gains from trade are fully exploited in a neoliberal
deal between states – γ is divided equally and fully. To model protectionism or other barriers
to full efficiency when the states are sovereign, let the gains available in peace be γp < γ, p for
“protectionism.” Now it is immediate that greater protectionism implies a tighter war constraint
and greater a∗, since there is now a greater economic efficiency motivation for war.62
Protectionist and colonial trade blocs in the 1930s were thought to have favored war, by in-
creasing the returns to acquiring territory where a major power could guarantee a flow of resources
and benefit from trade based on comparative advantage. Rosecrance and others have argued that
free trade favors peace because it provides states with a cheaper means of obtaining benefits that, in
example). See Copeland 2015, chap. 1, for a review of work on economic interdependence and war.61This is the same argument made above about democracies having less value for conquering additional
territory (µ), but applied here to one form of γ (gains from trade).62This can be seen directly from Figure 1; decreasing γp shifts the line giving the payoff for peace down
but has no effect on the value for war if, post-war, the winner can collect the full gains from trade γ.
35
a world with less free trade, might require conquest to obtain.63 The model formalizes and extends
this argument: Free trade favors lower costs of anarchy not because it is a benefit foregone while
fighting, but because it reduces the relative economic appeal of successful conquest. Thus freer
trade should allow states to maintain lower equilibrium arms levels. Lower costs of anarchy in turn
make a peaceful status quo more beneficial and so more robust against temptations for war.64
Differences in relative power and greed, and multipolarity
For clarity the baseline model assumed two states with equal resources, the same preferences over
territory, and the same war costs. I next consider a variant that allows for asymmetries, and briefly
discuss the restriction to two states.
Relative power and arms levels
Existing IR theory makes no predictions about how the military burdens of two adversaries
should be expected to vary with relative resources.65 If state A grows larger than state B, how will
force sizes or military spending change? Empirically, examples from Table 1 suggest a tendency
for the smaller state in a rivalrous dyad to make a larger military effort, relative to resource base.
Perhaps this makes intuitive sense – the smaller state needs to spend a larger fraction of income to
match the arms levels produced by an even smaller fraction spent by the larger state.
Analysis of an asymmetric version of the baseline model largely supports this intuition, but
with qualifications. In the online appendix I consider an extension in which the states have re-
63Rosecrance 1987.64Copeland’s “trade expectations” theory argues that a trading state that anticipates increased (extreme)
protectionism may go to war to avoid loss of gains from trade. This is a dynamic version of the model’slogic just described; Copeland 2015.
65By contrast, following Olson and Zeckhauser 1966 a large literature examines defense spending withinalliances.
36
sources r1 and r2 that they can spend on arms, where R = r1/r2 indexes state 1’s “relative power.”
The analysis identifies offsetting strategic dynamics. First, greater resources for state 1 means that
it can get a bigger advantage from mobilizing and attacking, which implies that a smaller state
2 needs a larger peacetime military burden to maintain deterrence. However, greater resources
for state 1 also means that state 2 has more to gain from successful conquest, a motivation that
presses the larger state to spend more to deter. The second strategic logic is referenced by Douglas
MacArthur’s comment that “Wars are caused by undefended wealth,” suggesting that rich countries
need to spend more to deter invasion.66
Nonetheless, further analysis shows that under a plausible assumption about how war costs
vary with relative size – namely, that the smaller state faces higher costs – the smaller state in a
rivalry will tend to have the larger military burden. Let i’s costs for war with j be cir j. In the case
of the ratio-form military technology, a closed-form solution for the lowest feasible equilibrium
military burdens, b∗i = a∗i /ri, can be derived. Figure 4 shows how state 1’s military burden varies
as we vary the ratio of its total resources to state 2’s (that is, R = r1/r2), for two levels of offensive
advantage and two levels of “greed.”67
The smaller state has the larger equilibrium military burden, and its burden increases as the
power disparity grows. The size of these effects increase with greed and offensive advantage,
and decrease with greater war costs. Note also that except in the case of high greed and high
offensive advantage, military burdens are quite insensitive to relative power ratios. This is due to
66The offsetting dynamics could help to explain why most states spend approximately the same share ofnational income on the military and have similar shares of population in the army, without a strong tendencyfor military burdens to vary with aggregate income.
67The parameters used for the figure are m equals .1 or .2, and µ − c equals .2 or .4 (greed net of warcosts). The closed-form solution is for the “Powell model” with γ = 0, so technically we need to assume asequential move extensive form. The same qualitative results hold for the model with γ > 0 in numericalexamples like those of Figure 4, but no closed form is possible. See online Appendix.
37
the offsetting strategic forces described above.68
Asymmetric “greed” or war costs
States can differ not only in relative size but also in the value they put on acquiring new
territory or costs of fighting relative to value for new territory. To extend possible asymmetries to
these factors, let µi ∈ [0, 1] be the weight state i puts on control of j’s territory, and ci be i’s costs
for war. We then have
Proposition 5. Let wi(ri, r j, µi, ci) be state i’s value for winning a war, and suppose that wi is
increasing in µi and decreasing in ci. Then for any military technology, increasing the value i puts
on control of j’s territory, or reducing i’s costs for fighting j, implies greater equilibrium military
burdens for both states.
The key point here is feedback effects – security externalities that work through increased
costs of deterrence. In an extreme case, a sufficiently greedy state could lead a status quo state –
one with zero intrinsic value for additional territory – to prefer war to eliminate the greedy regime
and so reduce the need for costly deterrence and competition. U.S. entry into World War II, and
arguably the Iraq War of 2003, were partly motivated by this logic, which accords more generally
with the idea behind Wilson’s notion of war “to make the world safe for democracy.”
More than two states
Waltz proposed to explain variation in major power war risk by the number of major powers.
He argued that more than two (“multipolarity”) implied higher risk from greater odds of miscalcu-
lation about who would fight with whom, due to complexity and incentives for “buck-passing.” But
he also argued that states balance, allying with the weaker side against a stronger aggressor. This
68With sufficient greed and extreme resource asymmetry, the bigger state prefers simply to attack the smallstate rather than pay the costs to deter invasion. This is why, for extreme imbalance, there is no peacefuldeterrent equilibrium in the high offensive advantage/high greed case in the figure.
38
tendency should create expectations of coalitions forming against aggressors under multipolarity,
thus some effect in the opposite direction.69
Multipolar balance of power models in which coalitions can form are complex, sensitive to
what seem like fairly arbitrary assumptions, and appear to support few general results.70 In ad-
dition, to date they abstract from features such as geography that have been critical influences on
what coalitions are likely to form in, say, European great power politics.71 These are reasons why
it makes sense for a baseline model to begin with relations between just two states. But it also
means that some important strategic dynamics cannot be addressed by the model, such as the U.S.
or Britain entering continental wars out of fear that victory by France or Germany would harm
them in the future.
Nonetheless, for an empirically common case of multipolar politics we can get a useful impli-
cation from the baseline model. If there is a “balancer” state that would intervene on behalf of state
j when state i attacks j, this amounts to defense dominance in the model: Either the war costs of
the attacker are increased, or the odds of victory are decreased (smaller m). By results summarized
in Propositions 3 and 5, in both instances the war constraint is relaxed and greater cooperation
becomes feasible between i and j.
The U.N. order and decolonization after World War II created a system with many small, poor,
formally independent countries, mainly in Africa and Asia. Invasions in the developing world with
the goal of conquest or annexation of territory have been extremely rare, in some part because
of developing states’ “recognition that they will probably meet strong Western opposition if they
embark on territorial aggression.”72 The implication from the model is then that these states can
69See Van Evera 1990, 36, for a good discussion.70Wagner 1986, 1993; Niou, Ordeshook, and Rose 1989; Powell 1999; Krainin 2014.71Models typically assume that any state can equally well attack any other state (conditional on relative
resources), whereas in reality spatial arrangements matter a great deal.72Zacher 2001, 242. See also Fazal 2007 and Jackson and Rosberg 1982, who argue that African “juridi-
39
afford to have smaller militaries than they otherwise might. Indeed, for many countries that do
not have a significant nationalist land dispute with any neighbors, the size of their military may
not reflect interstate deterrence at all, but instead is a matter of domestic demand (aD as discussed
above). In other words, for large parts of the world, there is no binding war constraint on arms
levels and international cooperation. The costs of interstate anarchy are low for these states as a
result of major power policies.
Conclusion
The model of interstate politics analyzed above may be summarized as follows. States can use
military effort to change the world beyond their borders in directions they like and also to invade
other states or defend against invasion. But because military effort is costly (there is a guns-butter
tradeoff), they face a strategic problem akin to repeated Prisoners’ Dilemma (PD). Threats or use
of force can gain short-run advantage on various issues, but if other states do the same they end
up spending more for no long-run gain. Thus agreements to limit costly competition by strategies
of conditional cooperation (e.g., Tit for Tat), often facilitated by international institutions, make
sense, even for highly “greedy” states.
However, because the strategic problem states face in anarchy is not a standard repeated PD,
this “neoliberal institutionalist” mechanism cannot necessarily be used to reduce the costs of an-
archy to zero. Unlike in repeated PD, a cooperating state can be eliminated or permanently disad-
vantaged (made unable to retaliate effectively) if the other state “defects.” As seen in the baseline
model, the extent of international cooperation is limited by states’ need to maintain enough arms to
deter mobilization and attack by potential adversaries. This “war constraint” sets a lower bound on
the costs of anarchy. In the model, the lower bound is determined by the offense-defense balance,
cal” states coordinated expectations of joint opposition to territorial aggression (balancing) among them-selves, so lowering the need for larger military capabilities.
40
the intrinsic value both states put on controlling additional territory, and the extent of gains from
cooperation on issues beyond homeland territory.
The model shows how diverse and apparently conflicting theoretical arguments can be syn-
thesized in the sense of being derived in a clear, connected fashion from a small set of Realist
premises. Some of its main implications are similar to those of the closest thing to a baseline
model in the literature, the two-type security dilemma theory posed by Jervis and developed by
Glaser, Kydd, and others.73 However, its core strategic logic is different. In this way it suggests an
alternative theoretical foundation.
In the model, the costs due ultimately to anarchy arise not from states’ uncertainty about
others’ territorial aims, but instead from a strategic dynamic rooted in political economy. For
instance, state A’s revisionism forces B to spend more to deter, which lowers B’s value for the
status quo, forcing A to spend more to deter B, which lowers A’s value for the status quo further,
and so on. Greater equilibrium military competition makes states more willing to risk war in
crises or try preventive war to avoid oppressive defense burdens in the future.74 Greater offensive
advantage in geography or military technology means that states have to arm more to get the same
degree of deterrence, thus endogenously lowering states’ values for living with the status quo and
exacerbating a “security dilemma” as this theory would understand it.
The standard two-type model has no way to understand tragic inefficiency between states with
some revisionist preferences.75 Here, by contrast, anarchy can cause tragedy for any set of states
not all known to be pure security seekers, due to the security externalities of costly competition in
73Jervis 1976, 1978; Glaser 2010; Kydd 2005.74For empirical examples from the lead ups to the World Wars, see McDonald 2011, Ferguson 1994,
Herrmann 1996, 228, Maiolo 2010, and Copeland 2000.75Recall that Jervis 1976 distinguished between “real” and “illusory incompatibility” and saw only the
latter as inefficient, a tradition carried on in the way that standard security dilemma arguments are presented.
41
arms or other military effort.76
Extensions to the complete-information model might introduce incomplete information about
preferences for territory (µ). Its absence is not intended to suggest that uncertainty about motiva-
tions is unimportant for interstate interactions. Jervis’ and much work following him demonstrate
otherwise.77 Starting with a complete-information baseline is useful for multiple reasons, including
clarifying the sources of the costs of anarchy and allowing us to understand just what incomplete
information about motivation would add.
Moreover, not all interstate competition is driven by uncertainty about the other side’s intrinsic
preference for territorial expansion. For instance, it seems plausible that Chinese and U.S. leaders
understand perfectly well that the former would like to revise the “territorial” status quo in the East
and South China seas, and that the U.S. would prefer that they did not. Likewise, the leaders of
North Korea and Iran correctly view U.S. administrations as having revisionist preferences con-
cerning their regime type, so that the U.S. can’t easily commit not to support avowedly democratic
challengers in these states in the event of domestic uprisings (as in Libya). This known U.S. revi-
sionism can then add to the incentive to develop nuclear arms.78 Such interactions may not be best
analyzed in terms of status quo types’ uncertainty about whether the other is revisionist, and yet –
by the strategic logic developed above – they can still be very costly and inefficient as a result of
anarchy.
76Constructivists hold that anarchy need not be costly because states could understand themselves not tohave any disagreements (such as revisionist territorial or issue preferences). No disagreement implies nocommitment problems. As noted, Realists and neoliberals might agree but question the relevance of theobservation. Still, this is not to say that changes in ideas and discourse are irrelevant; they can have largeeffects on conceptions of self-interest and conflict of interest. For example, the germ theory of disease wasan intellectual innovation that led, correctly, to increased perception of common interest across states insome areas (Cooper, 1989). Smithian and Ricardian theories of international trade are another importantexample (e.g., Morrison, 2012).
77Jervis 1976, 1989.78See Fearon 2011 for more on these examples.
42
Even though the costs of anarchy have been at the center of arguments in International Re-
lations for years, there have been almost no efforts to measure them empirically, beyond studies
of patterns of interstate war. The analysis here suggests that arms levels provide a theoretically
defensible measure, because failures of cooperation in areas where repeated interactions could in
principle support deals should not necessarily be blamed on anarchy. Moreover, arms levels are a
convenient measure both for gauging welfare consequences of anarchy across time and space and
for assessing how much a “third image” logic influences these costs. Regarding welfare conse-
quences, arguably a couple percent of GDP for most countries these days – the median was about
1.5% in 2012 – is not that much tragedy, especially if one considers that for many countries, some
large fraction of this is wholly about domestic stability or military politics rather than interstate
competition. An important point – consistent with “defensive realism” and critics who stress do-
mestic determinants of territorial preferences – is that very low costs of anarchy in a nuclear world
with many industrialized, free trading, nationalism-satisfied major-power democracies is implied
by a Realist baseline model.79
If antitrust law permitted it, profit-seeking firms would merge to form a monopoly. Likewise,
the two states in the model could in principle completely eliminate the costs of anarchy by merging
or federating, after which they could end military efforts directed at external defense. The condi-
tion of anarchy among states is not a natural given but instead an endogenous choice made by state
leaders. States do not merge for the same two main reasons that combatants in civil wars find it so
difficult to agree on powersharing arrangements. First, it is risky to trust that a new government’s
paper constitution will protect one from attack and expropriation after one has disarmed or ceded
control to a powerful higher-level military force. Second, in the modern period when nationalist
sentiment can be very strong, a leader and his or her constituents may simply have a large “con-
79Note that low military burdens in recent years are something to be explained. In earlier periods govern-ments spent almost all of what they could raise through taxes on the military.
43
sumption value” for formal independence, so that the costs of anarchy are a price of nationalist
satisfaction. The higher the price, the stronger the incentive for small states to merge and for major
powers to seek stronger institutions at international or regional levels. But strong international
institutions will be harder to build due to the underlying conditions that create the need for costly
deterrence. When the costs of anarchy are low (for example, at present, for the reasons suggested
above), genuine pooling of sovereignty might be more feasible but there is also less call for it, and
more scope for international institutions and conditional cooperation between independent states.80
80I consider issues that arise when we make anarchy an endogenous choice in work in progress.
44
References
Acemoglu, Daron, and James Robinson. 2006. Economic Origins of Dictatorship and Democracy.New York: Cambridge University Press.
Axelrod, Robert. 1984. The Evolution of Cooperation. New York: Basic Books.
Baliga, Sandeep, and Tomas Sjostrom. 2004. Arms Races and Negotiations. Review of EconomicStudies 71:351–369.
———. 2011. Conflict Games with Payoff Uncertainty. Ms., Northwestern University and RutgersUniversity.
Boix, Carles. 2003. Democracy and Redistribution. Cambridge: Cambridge University Press.
Brooks, Stephen G. 1997. Dueling Realisms. International Organization 51 (3):445–77.
Bueno de Mesquita, Bruce, James Morrow, Randolph Siverson, and Alastair Smith. 2003. TheLogic of Political Survival. Cambridge, MA: MIT Press.
Coe, Andrew J. 2012. Costly Peace: A New Rationalist Explanation for War. Ms., University ofSouthern California.
———. 2015. The Modern Economic Peace. Ms., University of Southern California.
Cooper, Richard N. 1989. International Cooperation in Public Health as Prologue to Macroeco-nomic Cooperation. In Can Nations Agree? Issues in International Economic Cooperation,edited by Richard N. Cooper, 178–254. Washington, DC: Brookings Institution.
Copeland, Dale C. 2000. The Origins of Major War. Ithaca: Cornell University Press.
———. 2015. Economic Interdependence and War. Princeton: Princeton University Press.
Davis, James W., Bernard I. Finel, Stacie E. Goddard, Stephen Van Evera, Charles L. Glaser, andChaim Kaufmann. 1998-1999. Correspondence: Taking Offense at Offense-Defense Theory.International Security 23 (3):179–206.
Debs, Alexandre, and Nuno P. Monteiro. 2014. Known Unknowns: Power Shifts, Uncertainty, andWar. International Organization 68 (1):1–31.
Downs, George W., and David M. Rocke. 1990. Tacit Bargaining, Arms Races, and Arms Control.Ann Arbor, MI: University of Michigan Press.
Fazal, Tanisha M. 2007. State Death: The Politics and Geography of Conquest, Occupation, andAnnexation. Princeton: Princeton University Press.
Fearon, James D. 1995. Rationalist Explanations for War. International Organization 49 (3):379–414.
45
———. 2010. Arming and Arms Races. Prepared for delivery at the 2010 Annual Meeting of theAmerican Political Science Association, September 2-5, 2010.
———. 2011. Two States, Two Types, Two Actions. Security Studies 20 (3):431–440.
———. 2012. A Simple Political Economy of Relations among Democracies and Autocracies.Unpublished ms., Stanford University. Presented at the 2008 Annual Meetings of the AmericanPolitical Science Association, Boston, MA.
———. 2015. Cooperation, Conflict, and the Costs of Anarchy. Presented at the 2015 AnnualMeetings of the American Political Science Association, San Francisco.
Ferguson, Niall. 1994. Public Finance and National Security: the Domestic Origins of the FirstWorld War Revisited. Past & Present (142):141–168.
Garfinkel, Michelle R., and Stergios Skaperdas. 2007. The Economics of Conflict: An Overview.In Handbook of Defense Economics, vol. 2, edited by Todd Sandler and Keith Hartley, chap. 22.Elsevier.
Gartzke, Erik. 2007. The Capitalist Peace. American Journal of Political Science 51 (1):166–191.
Gat, Azar. 2005. The Democratic Peace Theory Reframed: The Impact of Modernity. WorldPolitics 58 (1):73–100.
Gavin, Francis J. 2012. Nuclear Statecraft: History and Strategy in America’s Atomic Age. Ithaca,NY: Cornell University Press.
Glaser, Charles L. 1992. Political Consequences of Military Strategy: Expanding and Refining theSpiral and Deterrence Models. World Politics 44 (4):497–538.
———. 1997. The Security Dilemma Revisited. World Politics 50 (1).
———. 2010. Rational Theory of International Politics. Princeton: Princeton University Press.
Glaser, Charles L., and Chaim Kaufmann. 1998. What is the Offense-Defense Balance and CanWe Measure It? International Security 22 (4):44–82.
Grieco, Joseph. 1988. Anarchy and the Limits of Cooperation: A Realist Critique of the NewestLiberal Institutionalism. International Organization 42 (3):485–507.
Herrmann, David G. 1996. The Arming of Europe and the Making of the First World War. Prince-ton, N.J.: Princeton University Press.
Hirshleifer, Jack. 1989. Conflict and Rent-seeking Success Functions: Ratio and Difference Mod-els of Relative Success. Public Choice 63:101–112.
———. 1995. Anarchy and Its Breakdown. Journal of Political Economy 103 (1):26–52.
46
Isard, Walter, and Charles H Anderton. 1985. Arms race models: A survey and synthesis. ConflictManagement and Peace Science 8 (2):27–98.
Jackson, Matthew O., and Massimo Morelli. 2007. Political Bias and War. American EconomicReview 97 (4):1353–1373.
———. 2009. Strategic Militarization, Deterrence and Wars. Quarterly Journal of Political Sci-ence 4 (4):279–313.
Jackson, Robert H., and Carl G. Rosberg. 1982. Why Africa’s Weak States Persist: The Empiricaland the Juridical in Statehood. World Politics 35:1–24.
Jervis, Robert. 1976. Perception and Misperception in International Politics. Princeton: PrincetonUniversity Press.
———. 1978. Cooperation Under the Security Dilemma. World Politics 30:167–214.
Jervis, Robert L. 1989. The Logic of Images in International Relations. New York: ColumbiaUniversity Press.
Kaysen, Carl. 1990. Is War Obsolete? A Review Essay. International Organization 14 (4):42–64.
Keohane, Robert O. 1984. After Hegemony. Princeton: Princeton University Press. Chapters 4-6.
Krainin, Colin. 2014. Multilateral War Bargaining. Quarterly Journal of Political Science 9:407–439.
Kydd, Andrew. 1997. Game Theory and the Spiral Model. World Politics 49 (3):371–400.
———. 2000. Arms Races and Arms Control: Modeling the Hawk Perspective. American Journalof Political Science 44 (2):228–44.
Kydd, Andrew H. 2005. Trust and Mistrust in International Relations. Princeton: PrincetonUniversity Press.
———. 2015. International Relations Theory: The Game-Theoretic Approach. New York: Cam-bridge University Press.
Liberman, Peter. 1998. Does Conquest Pay?: The Exploitation of Occupied Industrial Societies.Princeton: Princeton University Press.
Lieber, Keir. 2000. Grasping the Technological Peace: The Offense-Defense Balance and Interna-tional Security. International Security 25 (1):71–104.
Lieber, Keir A. 2007. The New History of World War I and What It Means for InternationalRelations Theory. International Security 32 (2):155–191.
47
Maiolo, Joseph. 2010. Cry Havoc: How the Arms Race Drove the World to War, 1931-1941. NewYork: Basic Books.
Martin, Philippe, Thierry Mayer, and Mathias Thoenig. 2008. Make Trade Not War? Review ofEconomic Studies 75:865–900.
McDonald, Patrick J. 2011. Complicating Commitment: Free Resources, Power Shifts, and theFiscal Politics of Preventive War. International Studies Quarterly 55 (4):1095–1120.
Mearsheimer, John J. 2001. The Tragedy of Great Power Politics. New York: Norton.
Meirowitz, Adam, and Anne E. Sartori. 2008. Strategic Uncertainty as a Cause of War. QuarterlyJournal of Political Science 3 (4):327–352.
Morrison, James Ashley. 2012. Before Hegemony: Adam Smith, American Independence, and theOrigins of the First Era of Globalization. International Organization 66 (3):395–428.
Morrow, James D. 1997. When Do Relative Gains Impede Trade? Journal of Conflict Resolution41 (1):12–37.
Niou, Emerson M.S., Peter C. Ordeshook, and Gregory F. Rose. 1989. The Balance of Power:Stability in International Systems. New York: Cambridge University Press.
Olson, Mancur, and Richard Zeckhauser. 1966. An Economic Theory of Alliances. The Review ofEconomics and Statistics 48 (3):266–279.
Onorato, Massimiliano Gaetano, Kenneth Scheve, and David Stasavage. 2014. Technology andthe Era of the Mass Army. Journal of Economic History 74 (2):449–481.
Oye, Kenneth A., ed. 1986. Cooperation Under Anarchy. Princeton University Press.
Padro i Miquel, Gerard, and Sylvain Chassang. 2007. Mutual Fears and Civil War. Ms., PrincetonUniversity and LSE.
———. 2010. Conflict and Deterrence under Strategic Risk. Quarterly Journal of Economics125 (4):1821–1858.
Posen, Barry R. 1993. The Security Dilemma and Ethnic Conflict. Survival 35:27–47.
Powell, Robert. 1991. Absolute and Relative Gains in International Relations Theory. AmericanPolitical Science Review 85 (4):1303–1320.
———. 1999. In the Shadow of Power. Princeton University Press.
———. 2006. War as a Commitment Problem. International Organization 60 (1):169–204.
Powell, Robert M. 1993. Guns, Butter, and Anarchy. American Political Science Review87 (1):115–132.
48
Reiter, Dan. 2003. Exploring the Bargaining Model of War. Perspectives on Politics 1 (1):27–43.
Rosecrance, Richard N. 1987. The Rise of the Trading State: Commerce and Conquest in theModern World. New York: Basic Books.
———. 1999. The Rise of the Virtual State: Wealth and Power in the Coming Century. New York:Basic Books.
Schatz, Arthur W. 1970. The Anglo-American Trade Agreement and Cordell Hull’s Search forPeace 1936-1938. The Journal of American History 57 (1):85–103.
Schweller, Randall L. 1996. Neorealism’s Status Quo Bias: What Security Dilemma? SecurityStudies 5 (3):90–121.
———. 1998. Deadly Imbalances: Tripolarity and Hitler’s Strategy of World Conquest. NewYork: Columbia University Press.
Sheehan, James J. 2008. Where Have All the Soldiers Gone? The Transformation of ModernEurope. Boston: Houghton Mifflin.
Singer, J. David. 1987. Reconstructing the Correlates of War Dataset on Material Capabilities ofStates, 1816-1985. International Interactions 14:115–32.
Skaperdas, Stergios. 1992. Conflict, Cooperation, and Power in the Absence of Property Rights.American Economic Review 82 (4):720–739.
Slantchev, Branislav L. 2011. Military Threats: the Costs of Coercion and the Price of Peace. NewYork: Cambridge University Press.
Snyder, Jack L. 1984. The Ideology of the Offensive: Military Decision Making and the Disastersof 1914. Ithaca, NY: Cornell University Press.
Thompson, William R., and David R. Dreyer. 2012. Handbook of International Rivalries, 1494-2010. Washington, DC: CQ Press.
Trachtenberg, Marc. 1999. A Constructed Peace: The Making of the European Settlement, 1945-1963. Princeton, NJ: Princeton University Press.
Van Evera, Stephen. 1984. The Cult of the Offensive and the Origins of the First World War.International Security 9 (1):58–107.
———. 1990. Primed for Peace: Europe after the Cold War. International Security 15 (3):7–57.
———. 2001. Causes of War: Power and the Roots of Conflict. Ithaca, NY: Cornell UniversityPress.
Wagner, R. Harrison. 1986. The Theory of Games and the Balance of Power. World Politics38 (4):546–576.
49
———. 1993. The Causes of Peace. In Stopping the Killing, edited by Roy A. Licklider. NewYork: New York University Press.
———. 2010. War and the State: The Theory of International Politics. Ann Arbor, MI: Universityof Michigan Press.
Waltz, Kenneth N. 1959. Man, State, and War. New York: Columbia University Press.
———. 1979. Theory of International Politics. Reading, MA: Addison-Wesley.
Wendt, Alexander. 1992. Anarchy is What States Make of It. International Organization46 (2):391–426.
Wohlforth, William C. 1999. The Stability of a Unipolar World. International Security 24 (1):5–41.
Zacher, Mark W. 2001. The Territorial Integrity Norm: International Boundaries and the Use ofForce. International Organization 55 (2):215–250.
50
Table 1: Top and bottom 15 states on two measures of arms levelsTop 15, military burden Top 15, armed forces/1000
Cold War post-Cold War Cold War post-Cold WarUSSR2,3 Kuwait1,0 Israel3,2 North Korea0,1
Oman1,0 Eritrea2,1 Taiwan1,0 Eritrea2,1
Angola0,2 Russia0,1 Jordan3,1 Israel3,2
Israel3,2 Oman0,0 North Korea0,1 Jordan2,0
China5,1 Saudi Arabia1,2 UAE0,0 Syria4,1
Vietnam2,2 Armenia1,0 Albania0,1 UAE0,0
Egypt1,7 Syria4,1 Bulgaria3,0 Greece1,1
Bulgaria3,0 Bosnia2,0 Mongolia0,0 Oman0,0
Iraq2,4 Liberia0,0 Greece3,1 Iraq1,4
Jordan3,1 Angola0,1 Iraq2,4 Taiwan1,0
Saudi Arabia3,4 Israel3,2 Laos0,0 Lebanon0,0
Syria4,3 Jordan2,0 Syria4,3 Singapore1,0
Hungary1,0 Tajikistan0,0 South Korea0,1 Libya1,1
Mongolia0,0 Yemen1,0 Turkey3,0 South Korea0,1
Djibouti0,0 Uzbekistan1,0 USSR2,3 Armenia1,0
Bottom 15, military burden Bottom 15, armed forces/1000Cold War post-Cold War Cold War post-Cold WarHaiti0,0 Costa Rica0,0 Gambia0,0 Haiti0,0
Mauritius0,0 Jamaica0,0 Niger0,0 Ghana1,0
Gambia0,0 Mexico0,0 Mauritius0,0 Niger0,0
Mexico0,0 Mauritius0,0 Burkina Faso1,0 Malawi1,0
Costa Rica0,1 Dominican Republic0,0 Lesotho0,0 Nigeria1,0
Niger0,0 East Timor0,0 Malawi2,0 Kenya1,1
Trinidad0,0 Trinidad0,0 Rwanda0,1 Burkina Faso0,0
Jamaica0,0 Ireland0,0 Sri Lanka0,0 Mauritius0,0
Sierra Leone0,0 Guatemala1,0 Kenya2,1 Papua New Guinea0,0
Panama0,0 Austria0,0 Benin0,0 Gambia0,0
Fiji0,0 Japan1,0 Papua New Guinea0,0 Mali0,0
DRC0,1 Malawi1,0 Ivory Coast1,0 Bangladesh0,0
Japan0,0 Papua New Guinea0,0 Jamaica0,0 Ivory Coast0,0
Nepal0,0 Ghana1,0 Mali1,0 East Timor0,0
Bangladesh0,0 Gambia0,0 Bangladesh0,0 Cen. Afr. Rep.0,0
Notes: First superscript is # of territorial rivalries in period; second is # states with a “po-sitional,” “ideological,” or “interventional” rivalry (Thompson and Dreyer 2012). Somehigh-spending states like North Korea are missing data on military burden and so do notappear in the military burden lists. Post-Cold War is after 1990.
51
Figure 1: Equilibria with and without stable deterrence
0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.5
1.0
1.5
Arms level a
Val
ue fo
r pe
ace
at a
vs
brea
k ou
t/war
●
value of peaceful sq at a: 1 − a + γ 2war payoff: p(1, a, m)(γ − c + 1 + µ)peaceful deterrent eqm exists at a*war, no deterrent eqm
a*
●
52
Figure 2: Territorial rivalry over time
1850 1900 1950 2000
01
23
4
year
terr
itoria
l riv
alrie
s pe
r st
ate
all statestop 15% on military capabs (COW CINC scores)
53
Figure 3: Peacetime armed forces per 1000 population
1850 1900 1950 2000
05
1015
20
arm
ed fo
rces
/100
0
U.S.A.
year1850 1900 1950 2000
46
810
1214
16
arm
ed fo
rces
/100
0
United Kingdom
year1850 1900 1950 2000
510
1520
arm
ed fo
rces
/100
0
France
year
1850 1900 1950 2000
24
68
1012
arm
ed fo
rces
/100
0
Germany
year1850 1900 1950 2000
46
810
1214
16
arm
ed fo
rces
/100
0
Italy
year1850 1900 1950 2000
510
1520
2530
arm
ed fo
rces
/100
0
Russia
year
1850 1900 1950 2000
510
1520
2530
35
arm
ed fo
rces
/100
0
Turkey
year1850 1900 1950 2000
12
34
56
arm
ed fo
rces
/100
0
China
year1850 1900 1950 2000
01
23
45
6
arm
ed fo
rces
/100
0
Japan
year
54
Figure 4: Relative power and military burdens
0.1 0.2 0.5 1.0 2.0 5.0 10.0
0.0
0.1
0.2
0.3
0.4
Ratio of state 1's economic size to state 2's (r1 r2, log scale)
Sta
te 1
's m
ilita
ry b
urde
n (a
1*/r
1)
0.0
0.1
0.2
0.3
0.4
●
● rel. off. adv., high greedrel. off. adv., low greedrel. def. adv., high greedrel. def. adv., low greedwar (no deterrent eqm)
55